Analysis of CFD Modeling Techniques over the MV-22 Tiltrotor

Jennifer Abras Robert Narducci Aerospace Engineer Boeing Associate Technical Fellow NAVAIR 4.3.2.1 The Boeing Company Patuxent River, MD 20670 Philadelphia, PA 19142 [email protected] [email protected]

ABSTRACT The V-22 Osprey is a tiltrotor aircraft designed to operate under a wide range of flight conditions. Its outer mold line geometry is aerodynamically complex in part because aerodynamic considerations were not primary influential factors for the major features of the aircraft. As mission requirements change and additional devices are added to the aircraft, questions regarding the aerodynamic impact must be answered. While many of these questions can be adequately answered using lower-fidelity methods, some situations require the use of higher-fidelity analysis. Computational fluid dynamics (CFD) is a tool that has been used frequently to answer aerodynamic questions associated with the V-22. However, the complexity of the aircraft makes this analysis challenging. Using unstructured grids is one way of reducing the lead time required to setup the simulation as unstructured grids lend themselves to modeling complex geometries. This paper compares independent OVERFLOW and FUN3D CFD analyses of the MV-22 tiltrotor in airplane mode over a range of angles of attack, and compares these results to data from a recent high-angle-of-attack wind tunnel test run in the Boeing V/STOL Wind Tunnel. The results lend insight into the choice of grid structure, near-body vortex generation, and numerical methodology, and reveal that care must be taken when setting up the CFD model as well as identifying any numerical phenomena that could be considered code specific.

Much of the early testing focused on the full- INTRODUCTION scale development (FSD) aircraft. The current MV-22 The V-22 Osprey (Figure 1) is a multi-service, Block B configuration has several shape changes multi-mission tiltrotor aircraft capable of operating in including nacelle and spinner; and fuselage sponson. many diverse flight conditions. It is designed to take-off Many avionic kits and antennae are housed in fairings that and land as a helicopter and fly like a turboprop airplane. protrude throughout the fuselage. Recently The Boeing During the design phase, the overall dimensions of the Company re-established its wind tunnel database for the aircraft were defined by shipboard compatibility MV-22 (Marines) and CV-22 (Air Force) variants in an requirements (for example to operate, fold and stow on an entry at the 20 FT x 20 FT Boeing V/STOL Wind Tunnel LHA-class ship; and to carry an F-404 engine (BVWT)4,5. A photograph of the MV-22 model is shown internally1). The wing thickness, sweep, and dihedral in Figure 2. were also determined by non-aerodynamic considerations2. Extensive wind-tunnel testing in the 1980s helped shape the wing-fuselage juncture, overwing fairing, rear fuselage upsweep, and tail configuration.

Figure 2. MV-22 BVWT 0468 high-angle-of-attack Figure 1. The V-22 Osprey3. 0.15-scale wind tunnel model. The V-22 Osprey is in full production and has flown many operational missions for the Marines in Iraq Presented at the American Helicopter Society 66th Annual and the Air Force in Afghanistan. As it’s capability is Forum, Phoenix, AZ, May 11-13, 2010. This is a work of the being verified and realized in fleet operations, the U.S. Government and is not subject to copyright protection in Marines and Air Force are actively persuing definition of the U.S. new missions to take advantage of it’s full potential. To APPROACH support new designs, conventional aerodynamic tools and Both unstructured (FUN3D) and structured handbook assessments are often used to estimate drag or (OVERFLOW) CFD approaches are compared and air loads and to size new parts. For aerodynamic benchmarked against new wind tunnel measurements interference effects, higher-fidelity analysis such as CFD obtained for the MV-22. is often utilized. Wind tunnel testing or flight test demonstrations are conducted whenever the level of uncertainty exceeds the ability of analytical means to FUN3D 10.8 predict accurately. However, these tests are expensive FUN3D9,10 is an unstructured CFD code and generally have schedule implications. developed at NASA Langley that solves the RANS equations using a node-centered 2nd-order upwind implicit Computational fluid dynamics (CFD) offers scheme. While there a variety of flux schemes and practical and timely analysis to address many turbulence models that are available, for this study the aerodynamic designs. Advances in computer technology Roe flux scheme and the Spalart-Allmaras turbulence and computational algorithms have allowed Reynolds- model have been selected, since this turbulence model is Averaged Navier-Stokes (RANS) solvers to be used available in both CFD codes studied. Additional relevant routinely in most aircraft programs. The V-22 Program is settings used within FUN3D consist of steady-state no exception and has relied on CFD for the past 15 years pseudo time-stepping formulation as well as mapped to provide results for the analysis of the many flow least-squares11. The mapped LSQ option is applied to regimes of the tiltrotor aircraft. Today, computational reduce errors seen in unweighted LSQ methods applied to models with tens of millions of grid cells are common and high aspect ratio elements near curved boundaries. It was lead to very accurate and sophisticated flow simulations. ultimately found not to have a significant impact on the CFD simulations for airplane-mode operations were results. studied, for example by Tai6, in the late 1990s. More recently, advances in CFD modeling to capture the OVERFLOW 2.1 rotation of rotor blades have allowed researchers to produce very sophisticated hover simulations. Potsdam OVERFLOW12 is also a RANS code developed studied rotor-fuselage interaction of the V-22 in hover at NASA Langley; however solutions are computed on capturing the asymmetric fountain effect7. Using a less structured overset grids. In this application of sophisticated rotor model Narducci et al. modeled the OVERFLOW, a 2nd-order finite-volume central interaction of two V-22 aircraft in shipboard operations8. differencing scheme with scalar dissipation is utilized with the Spalart-Allmaras turbulence model. Steady-state Although every attempt was made to optimize solutions are obtained using pseudo time-stepping. Hole the aerodynamics of the V-22, engineering development cuts and block-to-block communication is established involving design to improve shipboard suitability, using the x-ray technique established by Meakin13. While survivability, reliability and fleet operations have resulted this technique does not always provide optimal in aerodynamic compromises that caused loss in air- communication among the grid blocks, it does allow the vehicle performance. As a result, design changes were blocks to move relative to each other during a calculation, incurred during Engineering Manufacturing Development for example rotor blades, which has been useful in to regain flight envelope. For instance, external wing applications of OVERFLOW to other V-22 problems. fences, forebody strakes and vortex generators were added to increase low-speed and high-speed performance GRID envelope. Though CFD methods are capable of capturing these aerodynamic interactions, application of these The OML geometry used in this study accurately methods pose a challenge and proper care and modeling is captures all the major components of the V-22. This required; grids designed for specific flow conditions or includes the wings, fuselage, nacelles, and empennage as grid adaptation techniques are sometimes required. well as many of the smaller components of the MV-22B Marine version namely the wing fence, conversion This paper compares independent CFD analyses fairing, forebody strakes, VCD, ARRP, FLIR, AA47 of the MV-22 in airplane mode. Two CFD codes, a fairings, main landing gear door bumps, SATCOM antenna, windshield wiper fairing, landing lights, and structured code, OVERFLOW 2.1, and an unstructured code, FUN3D 10.8 are utilized to generate solutions over glide slope antenna. The surface is shown in Figure 3. a range of angles of attack for conditions corresponding to For FUN3D, the grid bounding this surface consists of mixed elements14, with prismatic layers near the 200-knot flight at 3000 feet. The rotors are not modeled boundaries and tetrahedral elements in the isotropic to simulate the unpowered airframe aerodynamic wind regions. The underlying surface geometry used to tunnel testing. Forces and moments are compared and generate the grid consists of the same PLOT3D surfaces benchmarked against the latest wind-tunnel used to generate the OVERFLOW grid. These surfaces measurements. Interactional aerodynamics and off-body were converted to NURBS surfaces for unstructured grid flow features are also compared. generation. The grid has 15.1M nodes and 55M cells. Table 2. Model-scale flight conditions. The viscous layering was set such that y+ is less than or equal to 1.0. However, when running an 8° prediction, Characteristic Value the maximum y+ was found to be 0.64 at all surface Mach 0.2 locations. An initial tetrahedral grid has been generated Angle of Attack 0°, 4°-20°Δ=2° using GridTool and VGRID. FUN3D has been used to Altitude Sea Level merge the elements in the boundary layer region to create Nacelle Airplane Mode prismatic layers, with pyramids inserted to transition to Flap and Elevator 0° the tetrahedral portion of the domain. Air Properties Standard Reynolds Number 1.5M per ft

RESULTS MODEL SCALE The model-scale (MS) CFD study was designed to benchmark FUN3D and OVERFLOW against a particular wind tunnel (WT) configuration which closely Forebody resembles the production MV-22 configuration. The Strake same grid is used for this case as for the full-scale (FS) study. Scaling to the model problem is obtained by VCD adjusting the Reynolds number. This case is presented purly as a force and moment comparison study; a more thorough look at flow physics will be presented using results from full-scale calculations. Figure 3. Surface geometry of the MV-22 computational model. This inital valdation study primarily compares OVERFLOW uses a similar computational the total lift, drag, and pitching moment on the aircraft as surface, but this grid uses a series of 59 overlapped grids. a function of angle of attack. The lift coefficient The total number of grid points is 44.4M. Many of the predictions, plotted in Figure 4, match closely between smaller parts, particularly the forebody strakes and the the two CFD codes. However, when compared to the SATCOM antenna, were a significant challenge to grid wind tunnel data the results indicate the presence of three and lead to cells of questionable quality. An unstructured distinct regions, the low angle of attack region, the mid grid approach is ideally suited for this challenging angle of attack region and the high angle of attack region. configuration. The low angle of attack region matches closely between FUN3D, OVERFLOW, and the wind tunnel data. However, as the angle of attack exceeds 8°, CFD analyses CONDITIONS predict that the flow over portions of the aircraft begins to Two sets of flight conditions were selected for separate, leading to more premature severe stall beyond this study. The first, provided in Table 1, represents a full- 16°. scale cruise flight condition. The second set of conditions in Table 2 was selected specifically to match conditions in One significant difference between the CFD the wind tunnel. The model scale is 15%. geometry and the WT model is the presence of vortex generators (VGs) which span the entire wing including Table 1. Full-scale flight conditions. over the fuselage. The VGs are known to delay wing separation, but are difficult to model in CFD and have Characteristic Value hence been omitted. The VGs are most likely the primary Mach 0.3 contributors to the differences in lift beyond 8° in angle of Angle of Attack 0°, 4°-20°Δ=2° attack. However, since adding the VG’s to an Altitude 3000 ft unstructured grid is not as difficult, these components Nacelle Airplane Mode may be added in the future in order to prove this pont Flaps and Elevator 0° difinitively. Another difference is that while the WT data Air Properties Standard were corrected for the sting, CFD results on other Reynolds Number 2M per ft configurations have confirmed that modeling the sting can 15 improve the correlations between CFD and WT data .

the FUN3D and OVERFLOW results for higher angles of attack. This difference is due in part to different predictions of the forbody/wing interaction between the two codes. This is studied more thorougly in the full- scale results section.

Wind Tunnel

CL FUN3D Overflow

Wind Tunnel FUN3D Overflow CM

0102030 Angle of Attack [deg]

Figure 4. Computed MS aircraft lift vs. WT data. The drag predictions plotted in Figure 5 match 0102030 closely, particularly between FUN3D and the WT measurements. OVERFLOW predicts slightly higher drag Angle of Attack [deg] which has been traced to the viscous component. However, even with the offset both predictions are close to the wind tunnel data in character and magnitude. This Figure 6. Computed MS aircraft pitching moment vs. data does not exhibit the same three-region division seen WT data. in the lift coefficient. The comparison of FUN3D and OVERFLOW to experimentally measured forces and moments reveal that predictions at angles of attack of 8° and below can be Wind Tunnel FUN3D made with confidence. Differences in lift beyond 8° are Overflow attibuted to the VGs which remain an important challenge to model computationally. Still other differences exist and are most visible in the the pitching moment comparison. Of greater interest to the V-22 program are calculations made at full-scale conditions. Further studies comparing FUN3D and OVERFLOW are made at this scale, with

CD comparisons to WT data used sparingly to provide a point of reference.

FULL SCALE Even by disregarding the rotor, there are many aerodynamic phenomena associated with vortex generation and vortex interaction for the V-22. These features change significantly as a function of the aircraft 0102030angle of attack. Figure 7 shows the isosurface of vorticity Angle of Attack [deg] plotted over the entire aircraft at an angle of attack of 6°. Rotational flow over the nacelles, VCDs, and forebody strakes is easily identified in this image. Of these, the Figure 5. Computed MS aircraft drag vs. WT data. forebody strakes and the VCDs are specifically intended to promote flow interaction under specific flight The predicted pitching moment plotted in Figure conditions. Under the conditions shown here the 6 has the same divisions seen in the lift data. Here, the forebody strake vortex is interacting directly with the root aerodynamics of the horizontal tail are strongly emphasized because of the large moment arm between the of the wing. Physically, this is the expected result and the computational model is able to capture this vortex. moment center and the horizontal tail. In this case, unlike the lift and the drag, there is a clear difference between Wing Fence Nacelle

Forebody Strake CL

VCD Windshield Wiper Shield

FUN3D Figure 7. Representative FUN3D vorticity isosurface Overflow colored by pressure at an angle of attack of 6°. Wind Tunnel Though there is no measured data regarding the strength or location of the predicted vortices, the 0102030 interaction, or lack thereof, of the flow with the various Angle of Attack [deg] components of the aircraft does influence the force and moment computations. Figure 8 through Figure 10 compare the computed lift, drag, and pitching moment Figure 8. Computed FS aircraft lift vs. WT data. results from FUN3D and OVERFLOW. The discussion The drag coefficient is plotted in Figure 9. Here which follows supports the argument that differences in OVERFLOW demonstrates a greater sensitivity in drag capturing the vortex interactions contributes significantly between the full-scale and model-scale conditions relative to the differences in the integrated forces and moments. to the FUN3D results. The OVERFLOW differences are Also included in the figures are data from the WT test to entirely due to skin friction. Like the lift coefficient, the provide an anchor. The WT Mach and Reynolds numbers predictions in the mid angle of attack region show some do not match those of the calculations; therefore, a precise difference between the two CFD codes. As before, this match in forces and moments is not expected; however, difference is likely because of differences in flow the data will provide reasonable trends. Comparison with separation predictions between the two codes. model-scale CFD results to WT data are addressed in the previous section. FUN3D The aircraft lift coefficient, plotted in Figure 8 Overflow may be divided into the same three distinct regions Wind Tunnel predicted in the MS analyses. The low angle of attack data spans a range of 0° to 8°. In this region the computed lift matches closely between the CFD codes. Unlike the model-scale results, the mid angle of attack

region between 8° and 16° contains larger differences. CD FUN3D shows a larger sensitivity to either or both of the Mach and Reynolds number changes. The full-scale FUN3D results indicate a propensity for greater attached flow through this range than indicated by the OVERFLOW results. The reasons for this divergence over the mid angle of attack range will be the topic of discussion later in this section. As before, angles of attack of greater than 16° exhibit more severe aircraft stall. As noted earlier, the CFD models do not account for 0102030 the effects of the VGs which tend to promote attach flow Angle of Attack [deg] over the relatively thick wing. Figure 9. Computed FS aircraft drag vs. WT data. For conditions which produce separated flow, steady-state solution methodology can produce misleading results as time-dependent flow phenomena are lost. As expected, however, the use of time-accurate solution methodology does not appear to play any significant role in the accuracy of the predictions at angles between both the subtle and significant differences of attack below 10°, where flow is generally attached. revealed by these plots. The most noticible difference is Table 3 compares the steady-state and time-accurate the increase in nose down pitching moment seen in the results obtained when running FUN3D using a percent OVERFLOW solution. The FUN3D predictions indicate difference computation between the time-accurate and the the initiation of the same phenomena at 10°, but does not steady-state solutions. These particular solutions were continue the same trend past this point until 18° is obtained using a full tetrahedral grid and different inputs reached. for the numerical solution methodology. At 10° the predicted time-accurate solution has reached a steady- The plot of the horizontal tail lift coefficient in state and matches closely when a steady-state solution is Figure 11 confirms that it is the differences in the forced. However, after 16° the flow over the wing has horizontal tail lift predictions that are causing the increase become more separated and the use of time-accurate in nose down pitching moment. This is not unexpected methods becomes more important since the unsteadiness since it is the horizontal tail that contributes significantly of the shed flow is of interest. In this case the average of to the control of the aircraft pitch. Just as in Figure 10 the the time-accurate solution at 18° does not exactly match FUN3D and the OVERFLOW predictions match through the steady-state solution. This difference is because the 10°. This includes a slight increase in the horizontal tail time-accurate prediction is able to model unsteady lift at 10°. After this angle the FUN3D results follow a phenomena at high angles of attack that the steady-state more linear trend with angle of attack, whereas the solution cannot. OVERFLOW data continues to increase in a somewhat nonlinear fashion. This continues until the horizontal tail Table 3. Steady-state vs. time-accurate CFD. stalls. AoA CL CD CM 10° 0.079% 0.053% 0.598% L 18° 4.104% 0.770% 0.115%

The pitching moment coefficient data, plotted in Figure 10, may be divided into the same three distinct regions. The physics that define these three regions are FUN3D Horizontal Tail C Tail Horizontal dominated by the interaction of the upstream flow over Overflow the wing with the horizontal tail. This interaction changes the local angle of attack at the tail and thus changes the 0102030 Angle of Attack [deg] lift generated by the tail.

FUN3D Figure 11. Comparison of horizontal tail lift. Overflow Wind Tunnel This discussion brings up the question of what upstream flow phenomenon is causing this discrepency in the horizontal tail lift. Looking for correlations in the flow phenomena reveals that it is the forebody strake vortex interaction with the wing that is the primary factor. First, the more linear of the two solutions is analyzed. CM This is done in an effort to identify what physical phenomena contribute to the improved slope prediction

First, the streamlines that describe the path taken by the vortex shed by the forebody strake are plotted. Figure 12 shows the FUN3D predictions for an angle of attack sweep. At angles of attack less than 6°, the vortex 0102030from the strake passes below the wing; at 8° it begins to interact with the wing and by 10° the interaction is quite Angle of Attack [deg] strong. The OVERFLOW solutions, shown in Figure 14 indicate a similar interaction with the wing; however, Figure 10. Computed FS aircraft pitching moment vs. differences between the solvers emerge with the flow WT data. surrounding the horizontal tail. It is clear from Figure 12 These line plots only show the overall that FUN3D predicts a strong influence of the wing wake predictions of the CFD solutions. A more detailed with the horizontal tail around 10°. On the other hand, the analysis of the flow around the aircraft reveals the reasons OVERFLOW solutions suggest that the wake passes above the horizontal tail at this angle. At angles above 4° 8°

10° 12°

14° 16°

18° 20°

Figure 12. Streamlines from the forebody strake for Figure 14. Streamlines from the forebody strake for an angle of attack sweep using FUN3D. an angle of attack sweep using OVERFLOW.

4° 8° 4° 8°

10° 12° 10° 12°

14° 16° 14° 16°

18° 20° 18° 20°

Figure 13. FUN3D oil line plots. Figure 15. OVERFLOW oil line plots.

10° the wake passes above the horizontal tail for both in the lift coefficient. This takes into account the larger predictions though in FUN3D this seems to be at a separation of the wing root flow in the OVERFLOW reduced distance. In the OVERFLOW solution the reason predictions balanced with the larger wing tip separation in that the flow is further from the tail is because of the the FUN3D prediction. Increasing to 12°, the larger region of separated flow predicted at the wing root. OVERFLOW lift drops slightly to account for the larger This region is illustrated by the jump in the streamlines wing root separation region that is not balanced by any near the trailing edge of the wing. significant growth in the FUN3D wing tip separation. This trend continues through 16° with the OVERFLOW From these streamline plots alone it is not root separation increasing, and the FUN3D wing tip completely clear what may be causing the increase in the separation increasing. Once the stall angle of attack horizontal tail lift in the OVERFLOW solution. In region is reached, both codes have massive separation general, this increase in lift would be caused by an over the wings. Though this is delayed by 2° in the increase in effective angle of attack seen by the horizontal OVERFLOW solution. tail between 10° and 18°. A region of separated flow upstream of the tail would contribute to this phenomenon The obvious asymmetry in the FUN3D solution by deflecting the flow away from the tail, creating a lower has the greatest impact on the rolling moment of the pressure region above the tail. This lower pressure region aircraft. Figure 16 plots the rolling moment coefficient of then causes the higher pressure flow from below the wing the two CFD codes compared to wind tunnel data. There to turn upward because of the local pressure gradient. is a greater prediction of rolling moment in the FUN3D This then turns the flow at the horizontal tail, increasing solution on the order of 1x10-3. While this is a small the local angle of attack and thus increaing the lift. This value, it is still larger than the OVERFLOW predictions train of cause and effect is better supported by plots of the which are closer to zero. The absolute magnitude of the separated flow regions explicitly using oil line plots wind tunnel data appears to lie somewhere in between which depict surface flow patterns. these two predictions, but is closer to the OVERFLOW prediction. Figure 15 plots the oil lines over the aircraft surface using the OVERFLOW solution for an angle of attack sweep. The separated region at the wing root begins at 10° just as it does in the FUN3D solution plotted in Figure 13. However, the size of this separated region progressively increases as the angle of attack is increased CRM FUN3D Overflow at a greater rate than the increase seen in the FUN3D Wind Tunnel solution. This is consistent with the previous argument that it is the flow separation at the wing root that is 0102030 influencing the tail lift the most. Angle of Attack [deg]

There is not only a difference in the pitching Figure 16. Predicted rolling moment compared to moment predictions plotted in Figure 10, there is also a wind tunnel data. difference in the lift and drag coefficients plotted in Figure 8 and Figure 9 in the mid angle of attack region. The asymmetry in the FUN3D solution has little However, the differences in these predictions are to do with the algorithms used in the CFD code, it is the reversed. The OVERFLOW solution is predicting lower grid that causes this asymmetry. The underlying surfaces lift than the FUN3D solution in this region, instead of used to generate this grid are the same PLOT3D surfaces higher lift as seen at the tail. Taking another look at the used to generate the OVERFLOW grid. These surfaces oil lines in Figure 13 and comparing these with similar oil are exactly symmetric about the baseline of the grid for all line plots in Figure 15 reveals that it is the flow over the components that exist on both the LHS and the RHS. wing that causes this difference in lift. Again, this is not Since the aircraft is inherently asymmetric because of the unexpected since the wing is the primary lifting surface of ARRP and the SATCOM an exact mirror of a semi-span the aircraft when in airplane mode. grid could not be generated. Therefore, the asymmetry must exist in either the unstructured surface mesh or in The similarities and differences in integrated lift the volume mesh. A direct comparison of the between FUN3D and OVERFLOW can be explained by unstructured surface mesh was undertaken by taking the comparing the flow over the wings in Figure 13 and mirror of the grid about the baseline axis. The two Figure 15. From low angle of attach to 10° the codes meshes were overlayed and compared to look for any agree well. At 4° there is only a small separation region significant differences around the tip of the wing and the at the tip of the left wing. This is not sufficient to nacelles. The only differences seen were in the exact significanly impact the lift coefficient. At 8° this locations of the nodes and element used to describe the separated region grows, accounting for the slight decrease surfaces. The surfaces themselves were exactly in lift coefficient. At 10° there is a small crossover point symmetric. The only other possibility is that the volume which is close to the mid-span of the wing. As the angle mesh is the reason for the asymmetry. Slices through the of attack increases the pressure distributions match volume did not reveal any notable differences. However, closely between the two codes until the separated flow, there may still be some key difference that is triggering predicted differently by each code, at the root and the tip early separation on the LHS of the grid. In an initial test of the wing begins to influence the wing mid-span. Over of the grid the conversion fairings were removed. Since the right wing this divergence occurs at 18°. these complex components were difficult to divide into patches there may have been some underlying differences that transferred to the volume mesh. Figure 17 compares the FUN3D solution with and without the conversion fairings. The solution on the modified grid is now symmetric over the wings at 4°. Since the PLOT3D surfaces were symmetric to begin with, this result supports that there was some asymmetry introduced into the surrounding volume grid by these components.

Modified Original

Figure 17. FUN3D oil lines comparing original and modified grids at 4°.

In an effort to reduce the asymmetric differences Figure 18. Comparison of pressure coefficient over the in the volume grid cells, a new grid has been generated by right wing at Y=160in. slicing the underlying VGRID restart file in half and generating a semi-span model of the RHS. This new Differences between the pressure distributions restart file was copied and mirrored about the mid-plane. occur at a lower angle of attack closer to the ends of the The two halves of the aircraft were stiched together so wing. Figure 19 plots the pressure coefficient at that every curve, every patch, and every source used to Y=100in, near the wing root. The predictions of both generate the grid is exactly symmetric. The two FUN3D and OVERFLOW begin to diverge around 10° at asymmetric surface components were reintroduced into this location. the grid independently. However, the background sources used to cluster the points on the ARRP and the SATCOM were copied over the symmetry plane. All the background sources in the grid have a field effect everywhere in the grid domain. Thus creating symmetric sources even if they are not necessary should in theory enhance the symmetry of the volume mesh. However, this procedure had little impact on the results. The symmetry was moderately altered, but the overall lift, drag, and pitching moment did not show any appreciable change. This result is positive when considering the lack of grid sensitivity for a given grid resolution on the solutions obtained, but is negative in the sense that it proves that at this grid resolution an approximatly symmetric result cannot be obtained on an asymmetric configuration.

While gridding more complex surfaces can present difficulties for both structured and unstructured methodologies, in regions where the surfaces are easier to model the results obtained by FUN3D and OVERFLOW Figure 19. Comparison of pressure coefficient over the are nearly identical. Figure 18 compares the pressure right wing at Y=100in. coefficient plots for an angle of attack sweep at Y=160in, COMPUTATIONAL METRICS and flow conditions. OVERFLOW, FUN3D and experimental data departed under conditions featuring Relevant computational metrics are provided in strong vortex interactions and large separated flow Table 4. These metrics take into account various factors regions. At mid to high angles of attack the experimental including grid size, number of iterations, machine type, data suggests that attached flow is aided by main wing and number of processors used. The criteria used to VGs. Both OVERFLOW and FUN3D separated over the define a converged solution isn’t precise. The FUN3D main wing with OVERFLOW showing a stronger solution was actually run to 6000 iterations, but 2000 was propencity to separate. chosen as the point where the lift, drag, and pitching moment coefficients appeared to stop changing when Separation over the wing initiates from the plotted against a reasonable scale. The last entry in the inboard section. This is also the region where the vortex table aims to eliminate this subjective factor from the from the forebody strake interacts strongly. The vortex is comparison by identifying the time per iteration. This captured differently between the two codes resulting entry also factors in the number of grid nodes and the ultimately in different integrated forces and moments. number of processors used. The machine used to OVERFLOW faces the challenge of maintaining accuracy compute the FUN3D results is a Dell PowerEdge M610 as the vortex passes among block boundaries where system with two quad-core Intel Nehalem processors per dissipation tends to be the largest. The effects of the node and 24 Gb RAM per node. The OVERFLOW vortex interaction with the wing are felt as far back as the results were computed on a HP BL460 G1 system with horizontal tail in the FUN3D solution. The OVERFLOW two Intel Xeon 5160 dual-core processors per node and solution shows a smaller interaction with the horizontal 12 Gb of RAM per node. Both systems used 64-bit tail. The solution symmetry was better maintained with processing. OVERFLOW than with FUN3D for reasons not fully understood, but currently under investigation. FUN3D Table 4. Computational metrics. asymmetry resulted in mixed separation regions at the Code FUN3D OVERFLOW wing tips which increased the predicted rolling moment. Processor speed 2.8 GHz 3.0 GHz Number of processes 128 40 For many V-22 applications, RANS methods Grid size [nodes] 15.1M 44.4M appear to be well suited to address difficult aerodynamic Number of iterations concerns. Ultimately however, careful evaluation of every 2000 8000 to converge 0° case solution is needed particularly for complex interactions Total wall time [hrs] 2.7 29.2 and separated flows. Time to solution/node 0.082 0.095 [sec/node] Future work on this project will attempt to Time/iteration/node identify the number of cells needed to approach grid 4.08x10-5 1.18x10-5 [sec/iter/node] independence as well as looking at the impact of modeling the wing and mid-wing VGs on wing separation, as well as studying unsteady effects at high CONCLUSIONS angles of attack and the effect of turbulence models on these unsteady effects. The culmination of this work will For complex aircraft care must be taken when be the analysis of the impact of the various aerodynamic modeling the flow. The MV-22 falls into this category, components on the stability and control of the aircraft. presenting challenges for RANS-based analysis. This paper presented independent solutions from FUN3D, an ACKNOWLEDGMENTS unstructured flow solver, and OVERFLOW, an overset structured flow solver. Care was taken to maintain The authors would like to acknowledge the equality between the geometry used in both grids. The resources provided by the HPC program, in particular the MV-22 airframe was analyzed at model- and full-scale in MHPCC, as well as the advice of the FUN3D airplane mode. Despite the lack of rotors, the development team, in particular Elizabeth Lee-Rausch aerodynamics of the problem is characterized by strong and Eric Nielsen. Wind tunnel data was provided by the vortex interactions. The 23% thick wing provides Boeing Company and was based on work performed additional challenges as separation occurs gently and under Boeing Independent Research and Development relatively early as angle of attack is increased unless VGs The authors would also like to acknowledge Jonathan are utilized. Chiew and Joseph Gillman for their contributions to the OVERFLOW solutions. This work was funded in part by Generally, reasonable agrement between FUN3D the V-22 Flight Dynamics Engineering lead Sean Roark. and OVERFLOW was achieved for flows which were The authors would also like to thank Steven Woods of attached and vortex interactions which were minimal. NAVAIR Flight Vehicle Performance for reviewing this Under these conditions CFD results also agreed well with paper. experimental data recorded for the same configuration REFERENCES 2001-2537, AIAA 15th Computational Fluid Dynamics

Conference, Anaheim, CA, June 2001. 1.McVeigh, M.A., Liu, J., O'Toole, S.J., and Woods, S.A., "V-22 Osprey Aerodynamic Development – A 14.Biedron, R., Vasta, V., and Atkins, H., “Simulation of Progress Review," 22nd European Rotorcraft Forum, Unsteady Flows Using an Unstructured Navier-Stokes Paper 2225, September 1996. Solver on Moving and Stationary Grids,” 23rd AIAA Applied Aerodynamics Conference, Ontario, Canada, 2.Rosenstein, H., and Clark R., "Aerodynamic June 6-9, 2005. Development of the V-22 Tiltrotor," AIAA-1986-2678, AIAA/AHS/ASEE, Aircraft Systems, Design and 15.Green. B., “Internal NAVAIR Communication.” E- Technology Meeting, Dayton, OH, October 20-22, mail interview, March 8, 2010. 1986. 3.Boeing Image Gallery, “http://www.boeing.com/rotorc- raft /military/v22/v22photos.html,” Last Accessed March 31, 2010.

4.Duffy, M., “BVWT 0467 - Wind Tunnel Test - V-22 Drag Reduction with Nacelle Sails, Optimum Flaps, and Vortex Control Device,” Boeing Document Number 901-909-664, November 2009.

5.Fox, E., “BVWT 0468 - 0.15 Scale V-22 High Angle of Attack Wind Tunnel Test Report,” Boeing Document Number 901-909-665, February 2010.

6.Tai, T.C., "Effect of External Components on V-22 Aircraft Forward-Flight Aerodynamics," Journal of Aircraft, vol. 37 no. 2, 2000, pp. 201-206. 7.Potsdam, M.A., and Strawn, R.C., “CFD Simulations of Tiltrotor Configurations in Hover,” American Helicopter Society 58th Annual Forum, Montreal Canada, June 11-13, 2002. 8.Narducci, R.P., Jiang, F., Liu, J., and Clark, R.W., "CFD Modeling of Tiltrotor Shipboard Aerodynamics With Rotor Wake Interactions," AIAA-2009-3857, 27th AIAA Applied Aerodynamics Conference, San Antonio, Texas, June 22-25, 2009. 9.Anderson, W. K., and Bonhaus, D. L., “An Implicit Upwind Algorithm for Computing Turbulent Flows on Unstructured Grids,” Computers and Fluids, vol. 23 no. 1, 1994, pp. 1-21. 10.FUN3D On-line Manual, “http://fun3d.larc.nasa.gov,” Last Accessed January 15, 2010. 11.Diskin, B., Thomas, J., Nielsen, E.J., Nishikawa, H., and White, J.A., “.Comparison of Node-Centered and Cell-Centered Unstructured Finite-Volume Discretizations. Part I: Viscous Fluxes," AIAA Paper 2009-0597. 12.Buning, P. G., Chan, W. M., Renze, K. J., Sondak, D. L., Chiu, I. T., and Slotnick, J. P., “OVERFLOW User’s Manual, Version 1.6ab,” NASA Ames Research Center, Moffett Field, CA, January 26, 1993. 13.Meakin, R., “Object X-Rays for Cutting Holes in Composite Overset Structured Grids,” AIAA AIAA-